P0–P2 polish: badges, architecture diagram, annotated segments, RU mirror, CI

[CreatmanCEO]

Summary

Single PR addressing the README/structure gap identified in deep comparison with the Hydrowatch standard. The actual tool was solid — install flow clean, VPS auto-focus from transcript is novel, cost auto-detection sensible. The README was understating the design (no architecture diagram, the cache-decoupling trick was invisible) and missing the social proof from the Habr article (9.3K reads).

What changed

P0 — credibility & accuracy

• Hero badges: License, Stars, Validate CI, Habr (9.3K reads), Claude Code Opus 4.7, bash + jq, platforms
• Hero blurb leads with social proof and the one-line value prop (pure bash + jq, no Node.js)
• Production screenshot embedded — docs/screenshots/statusline-live.png (220 KB) from the Habr article: 100-hour session, 8320 lines, $40 saved, three VPS at a glance
• Topics + description — applied via gh api after merge (next message)

P1 — structural strengthening

• docs/architecture.svg — cache-decoupling diagram. Three boxes with arrows: vps-poller.sh (daemon) → /tmp/vps-*.json (cache) → statusline.sh (fast read), plus the Claude Code transcript driving auto-focus. The architectural insight that makes a per-prompt statusline feasible.
• docs/segments-annotated.svg — per-segment breakdown with colour-coded labels above and below the rendered bar (model, git, lines, cost, 5-hour quota, weekly quota, duration, auto-focused VPS, context).
• New Architecture section with the diagram and a component-role table.
• Configuration table extended with the previously undocumented VPS_POLL_INTERVAL, VPS_SSH_TIMEOUT, VPS_STALE_SEC, COST_MODEL, TMUX_BRIDGE knobs.
• Troubleshooting gained VPS shows DOWN but server is fine and Auto-focus picks the wrong VPS entries.
• Limitations section: transcript-window heuristic, OS credential storage assumption, bash + jq required, global SSH timeout, no metric history, Windows-tmux unsupported, daemon best-effort.
• Related cross-link to claude-code-antiregression-setup and ai-context-hierarchy (sister repos, same author).
• CLAUDE.md for this repo — Level 1 example documenting architecture, CRITICAL RULES, commands, patterns. Pairs with the ai-context-hierarchy pattern.
• CHANGELOG.md (Keep a Changelog) starting at 0.1.0 → 0.2.0.
• CONTRIBUTING.md with priority list (per-distribution packages, language locales, alternative VPS metric backends, Fish/Zsh, glyph theming, systemd-user unit).
• .github/workflows/validate.yml — bash -n syntax, ShellCheck warning level, all internal Markdown links resolve, presence of LICENSE/CHANGELOG, every docs/* asset referenced from README exists, SVG well-formed XML, statusline.conf sourceable.

P2 — content

• README.ru.md rewritten to mirror the new English structure. The Habr badge in Cyrillic, links to the Habr article as primary Russian social proof.
• Author signature expanded with Habr / dev.to profile links.

What is NOT in this PR

• Demo GIF / asciinema cast. The statusline-live.png from the article carries the value proposition; a recorded GIF is a separate task per user instruction.
• Submission to awesome-claude-code. That channel is currently blocked for non-collaborators across the maintainer's repo; deferred until they re-open.
• PR to AndyShaman/claude-statusline with a back-link. Tracked separately.

Test plan

• bash -n clean on statusline.sh, vps-poller.sh, install.sh, statusline.conf
• Both new SVG files are well-formed XML
• All docs/* assets referenced from README and README.ru.md exist on disk
• All internal Markdown links from README / README.ru.md / CHANGELOG / CONTRIBUTING / CLAUDE.md resolve
• LICENSE and CHANGELOG.md present
• ShellCheck (CI runs this — not installed locally)

🤖 Generated with Claude Code